DS DevShelfHub Projects · AI tools
Articles / Inside Manus 1.6: The AI Agent Meta Just Acquired — Browser Control, Mobile Builds, and Real Data Analysis

AI Engineering

Inside Manus 1.6: Browser Control, Mobile Builds, and a Tutorial

By DevShelfHub

A complete walkthrough of Manus 1.6 after the Meta acquisition — the new agent architecture and 1.6 Max model with 30% better deep-research persistence, My Browser driving your real Chrome session through Amazon, surgical design-view edits with the mark tool, building a Parallel Lives mobile app previewable on a real phone via Expo Go, and a real 90-day credit-card spending analysis exported to Excel.

Inside Manus 1.6: Browser Control, Mobile Builds, and a Tutorial

Introduction

Meta acquired Manus — the autonomous AI agent that had already been getting traction for its deep research and tool-use capabilities — and almost immediately the platform got an upgrade that makes the acquisition price tag easier to understand. Manus 1.6 ships a new agent architecture, the Manus 1.6 Max model with ~30% better persistence on deep research, plus the ability to drive your real browser, design visuals, build mobile apps, and run real data analysis end-to-end.

This walkthrough covers what Manus actually does, what changed in 1.6, and four real demos: a browser-driving shopping task on Amazon, a logo design workflow with surgical edits, building a working mobile app from a spec, and a credit-card spending analysis from a real bank export to a polished Excel report.

📚 Table of contents

  • What Manus actually is (and what makes it different)
  • What changed in Manus 1.6
  • Demo 1 — My Browser: driving your real Chrome session
  • Demo 2 — Design view with mark-and-edit refinements
  • Demo 3 — Building a mobile app from a spec
  • Demo 4 — Real bank-statement analysis to Excel
  • Why Meta acquired Manus
  • Common mistakes and best practices
  • Frequently asked questions

🤖 What Manus actually is

Manus is an autonomous AI agent — with execution, not just generation. Unlike a chat model that spits text, Manus can run code, hit APIs, automate workflows, and deliver finished outputs. The integrated tool surface includes:

  • Web browsing (now including your browser, with your sessions).
  • Shell access.
  • Data processing.
  • File creation and manipulation.
  • End-to-end research, build, test, deploy loops.

Early reputation was built on deep research. 1.6 is what made it interesting for general production workflows.

🆕 What changed in 1.6

🧠 Manus 1.6 Max

New agent architecture with significantly better one-shot performance and ~30% more persistence in deep-research tasks. Better tool calling under load.

📱 Full mobile app builds

Manus can now scaffold and ship iOS and Android apps end-to-end from a spec, and you can preview them on your phone via Expo Go.

🎨 Design view

A visual workspace for generating and editing images, logos, and creative assets — including surgical text and region edits.

🌐 My Browser

Drive your real Chrome session — with your existing logins — so tasks that require authentication just work.

🛒 Demo 1 — My Browser driving Amazon

The prompt: “Go to amazon.ae, find the best coffee that will be delivered today, give me three options, then wait for my approval and buy them.”

With My Browser enabled (under the Chrome icon), Manus opens a tab in your real Chrome session, authenticates against your existing Amazon login, navigates to the storefront, filters by same-day-delivery availability, scrolls the results, picks three contenders, and surfaces them to you. On approval, it adds the chosen item to your cart — without you touching the keyboard.

This is the feature that’s genuinely hard to do well, because most agent frameworks fight authentication and bot-detection. Driving your real browser sidesteps both.

🎨 Demo 2 — Logo design with surgical edits

Design view supports two key workflows beyond simple generation:

  • Comment + iterate — drop comments on any part of an image (“make this text more modern”) and apply all comments in one revision pass.
  • Mark tool — lasso a specific region of the image and prompt against just that region (“make this corner red”, “turn this curve angular”).

The workflow that actually works: generate 10 logo variations, pick the closest, then layer in text-edit + mark-edit refinements until the brief is satisfied. Beats round-tripping through a chat interface and re-prompting from scratch.

📱 Demo 3 — A mobile app from a spec

The brief: Parallel Lives — a global daily-question app where users answer one thoughtful question per day and see how their response compares to millions of others. Pasted in, told Manus to use the previously chosen logo, hit enter.

The output, after two follow-up prompts for layout tweaks:

  • Daily question UI with answer submission.
  • Insights tab with data visualisation across demographic groups.
  • Profile view, recent answers, world responses, filters by country.
  • Demo data seeded for first-run experience.

Preview workflow: scan a QR code, open Expo Go on your phone, and the app runs natively on your device. Haptics, smooth scroll, real touch feedback — the lot. The kind of app that previously took two years of part-time work, scaffolded in minutes. Polish still requires manual passes, but the skeleton is real and shippable.

📊 Demo 4 — Real spending analysis

Workflow: download 90 days of credit-card transactions from your bank as a Google Sheet, paste the sheet URL into Manus, then ask for “detailed spending analysis — categorise transactions, chart the categories, surface the largest expenses, and convert AED to USD.”

Manus opens the sheet, downloads it locally, runs analysis, and produces:

  • Spend totals (e.g. ~$17K over 90 days), per-category breakdowns, largest single transactions.
  • Daily and weekly spend charts, day-of-week distribution.
  • An executive summary, a zip with the full report, and the Python source it used.
  • On request, a clean Excel workbook with key findings, category spend, recurring transactions, and all transactions — ready to send to an accountant.

The interesting part isn’t any single chart — it’s that Manus produced the entire pipeline (open → analyse → visualise → export → reformat on demand) without leaving the agent loop.

🏢 Why Meta acquired Manus

Tool-rich autonomous agents are the next platform layer, and the leading candidates were limited: either strong on research but weak on execution, or vice versa. Manus had both, plus a UI shape that actually worked for non-developers. Browser access using your session, surgical design edits, mobile build outputs you can preview on a real device, real data analysis from real files — the feature set reads less like a chat product and more like a platform.

Manus runs on free credits at signup and a paid plan beyond that. With Meta’s distribution and capital behind it, expect tighter integration with Meta’s broader stack and a sustained push on enterprise-grade reliability.

✨ Best practices & common mistakes

✅ Do

  • Use My Browser for anything that needs authentication.
  • Write a structured brief in a doc, then paste it into Manus — one-liners produce toys.
  • Use the design mark tool for fine edits instead of regenerating from scratch.
  • Preview mobile builds on a real device via Expo Go — the desktop preview undersells the result.

❌ Don’t

  • Approve every browser action blindly — it’s your real Chrome session.
  • Treat the first generation as final. Plan two or three iteration passes.
  • Hand it sensitive financial files without scoping what it can do with them.
  • Expect App Store-ready polish out of one prompt. Skeleton, yes; production, not yet.

Explore More on DevShelf

Inside Manus 1.6: The AI Agent Meta Just Acquired — Browser Control, Mobile Builds, and Real Data Analysis FAQ

Is Manus free?

Free tier with credits at signup. Beyond that you’re on a paid plan; pricing depends on which model and how heavy the workflows are.

How safe is My Browser?

Manus asks for explicit authorisation when opening a new tab in your session, and you can revoke access at any time. Still — treat it like granting power to a contractor. Don’t leave purchase tasks running unattended.

Can the mobile build be exported as a real codebase?

Yes — you can view all the code, the database, files, and settings inside the workspace, and pull it down for further development.

Is Manus better than competitors like OpenClaw or Perplexity Computer?

Different strengths. OpenClaw is self-hosted with skill compounding; Perplexity Computer is cloud-native with parallel sub-agents. Manus wins on browser sessions and mobile-build polish.

What happens after the Meta acquisition?

Officially, business as usual with continued shipping. Expect deeper Meta-stack integration over time and a push toward enterprise reliability.